Real Temp - New temp program for Intel Core processors

[rge]

Well with Real Temp promoting controversy by reporting different temps, dispelling the myth of Tjmax based on mobile cpus, and getting closer to the truth with the nonlinear calibration, I think Real Temp is partly responsible for getting intel to see the light on temps, at least providing 45nm tjmax's, confirming what you already discovered regarding idle temps, and most importantly suggesting a future fix. So, Unclewebb, definite Big !

Interestingly, all the 45nm Quads you can add 29-30C to Tcase and get Tjunction. All the 45nm Duos you can add 26-28C to Tcase and get Tjunction. I know one of your 65nm did not fit that mold, but wonder if that was an intel mistake or intel farting around with one model. The higher the power the greater the gradient, which makes sense.

[Praz]

What annoys me most is that a company like Intel which supposedly has one of the best manufacturing processes around has made such a fundamental mistake like this.

As far as Intel is concerned there has been no mistake at all. The sensor from the beginning had a single purpose and it performs exactly to design specs. That it is unable to serve a second purpose that others wish for by no means compromises the original design.

[BeastNotro]

Well i should have also stated when i used the probe at full load it was pretty darn close within in 1-2C at base of heatsink from what realtemp was stating it was at.

So i would still say 95C is the TJ MAX for Q6600

[Ol'Baditude]

I agree with you Beast. My tests support this as well.

[BeastNotro]

The other thing is if someone has a sensor that is all out of whack .. say 3 cores read 60-58-57 and the one core reads .. say 42C .. well it does not take a genious to figure it out that one sensor is ON CRACK and can't run as fast as the rest sort of speak.

Now if one were reading 75C and the rest read 60-58-56 . then the one that is 75C is on SPEED and is running away form the rest to the fire brigrade, obviously it is not reporting correctly.

Now in the case of ME .. i had to cores reading 10C higher then the other two cores and that was before i lapped my Thermal Right Ultra 120 ... took 6 hours to get it nice and flat .... then i used ThermalRight Chillfactor which is still on today ... only applied a thin layer on the IHS (CPU CAP) and now my temps full load are within 62-58-57-57 .. where as before it was 72-70-60-58.

The thing is the cores will fluctuate like now this sec its 61-58-52-54. lol

[rge]

Given IR reading of E8400 casing is 95C at DTS=0, and now that intel has finally provided tjmax, we finally know the temp gradient is approximately 5C from tjunction to casing at idle, undervolted, at least in cases where there is a solder attach adhesive like E8xxx, (maybe slightly more for adhesive attach), though there are other variables which we can not account for completely. But given both E8400 and Q6600 has solder attach (solder attach is ~50W*M/K from same intel slides that show solder attach melting point at ~220C if memory serves correctly), I would guess same ~5C gradient for Q6600 exists as well, which would make it 100 tjmax, since IR casing was 95C at DTS=0. Same for E6850, etc. Saw Unclewebb was talking about changing many of those at 95 back to 100, would assume for that reason. The adhesive attach thermal conductivity is several times lower, so dont know if > 5C exists for those, even if all other variables same.

Edit: Also if Q6600 GO was 100, not only would it have similar 5C gradient die to casing temp at idle, undervolted state, but also the same 30C difference from Tcase max to tjunction max as all the 45nm quads, though granted that is a completely different nm process. So would the Q6600 B stepping then be 90 tjmax (5 C more than IR measurement at idle, undervolted, and 28C more than tcase max...

Well intel solved one small issue with 45nm, but clearly the arguing about 65nm tjmaxs will continue

[unclewebb]

Your right Praz. Intel has never wanted users using any information from these sensors to report absolute core temperatures and their announcement was more like, "You're all wrong. We don't know what your core temperature is but we do know that all these programs must be wrong." That doesn't help the enthusiast community one bit. A comparison against a calibrated diode would have been nice to see. The presentation was more of a sales job for Core i7.

Based on your testing rge, the 5C delta is hard to believe and for this delta to remain at exactly 5C from idle to TjMax doesn't really add up either. At TjMax I can believe that delta might exist but it's harder to accept that at idle.

I'm going to throw together a new version of RealTemp that uses Intel's new information and then we can do some testing to see if this seems reasonable. At TjMax=100C, my E8400 has a lot of error to cover up down low. The RealTemp calibration method might still be a valid way to learn how your sensors respond when far away from TjMax.

At least no one's complaining about the color of the future logo for my toggle button today.

[kpo6969]

RealTemp 2.74 vs CoreTemp 99.3
unclewebb, if this helps

[graysky]

I read the PP slidedeck as well, but there is no mention of xeon equivalents of the Q9000 series (i.e. my X3360). I'm assuming it too is 100 °C?

@unclewebb - I'm running the latest beta of RT, but it is defaulting to 95 for my X3360... did you add support for xeon CPU detection?

The other thing that pissed me off is that I saw no mention of the older 65nm cores.

[loonym]

I read the PP slidedeck as well, but there is no mention of xeon equivalents of the Q9000 series (i.e. my X3360). I'm assuming it too is 100 °C?

No reason it wouldn't be since they use the exact same stepping.

@unclewebb - I'm running the latest beta of RT, but it is defaulting to 95 for my X3360... did you add support for xeon CPU detection?

It will continue to default to 95 until a new version is released. Simple enough matter to adjust in settings.

[skinnee]

Q9450 = Tjmax of 100ºC.

Damn, means my temps are 5º higher...

[loonym]

Q9450 = Tjmax of 100ºC.

Damn, means my temps are 5º higher...

Also means you have a 5 degree higher ceiling til tjmax so no big deal either way really I think.

[Mr Roboto]

As far as Intel is concerned there has been no mistake at all. The sensor from the beginning had a single purpose and it performs exactly to design specs. That it is unable to serve a second purpose that others wish for by no means compromises the original design.

You're probably correct but it still seems a very lazy\indifferent approach to something that's important to enthusiasts as well as many others. Maybe I'm just being somewhat unfair. I don't know.

[jason4207]

Given IR reading of E8400 casing is 95C at DTS=0, and now that intel has finally provided tjmax, we finally know the temp gradient is approximately 5C from tjunction to casing at idle, undervolted, at least in cases where there is a solder attach adhesive like E8xxx, (maybe slightly more for adhesive attach), though there are other variables which we can not account for completely. But given both E8400 and Q6600 has solder attach (solder attach is ~50W*M/K from same intel slides that show solder attach melting point at ~220C if memory serves correctly), I would guess same ~5C gradient for Q6600 exists as well, which would make it 100 tjmax, since IR casing was 95C at DTS=0. Same for E6850, etc. Saw Unclewebb was talking about changing many of those at 95 back to 100, would assume for that reason. The adhesive attach thermal conductivity is several times lower, so dont know if > 5C exists for those, even if all other variables same.

Edit: Also if Q6600 GO was 100, not only would it have similar 5C gradient die to casing temp at idle, undervolted state, but also the same 30C difference from Tcase max to tjunction max as all the 45nm quads, though granted that is a completely different nm process. So would the Q6600 B stepping then be 90 tjmax (5 C more than IR measurement at idle, undervolted, and 28C more than tcase max...

Well intel solved one small issue with 45nm, but clearly the arguing about 65nm tjmaxs will continue

Exactly what I was going to say. If we take Intels 100* Tjmax as fact then we have to assume a 5* gradient based on the testing. If we can then assume that the gradient will be very similar on 65nm then the 65nm CPUs should be 100*C as well.

[unclewebb]

The question I'm curious about is how does Intel determine TjMax? Is it an accurately measured value or is it an assumed value? Do they have a device that can accurately measure the core and determine that it is at 100C or do they measure 95C on the IHS of a finished unit and then say something like, "the core is probably about 5C hotter so we'll add on 5C and call it TjMax=100C." That's a serious question and one we'll never know the answer of.

I guess I'm not 100% convinced that switching RealTemp to TjMax=100C is going to result in more accurate temperatures. There was more important information left out of the IDF presentation than included in it. The only thing I'm really happy about is that I didn't spend a $1000 + travel to get my hands on that presentation. My rant for the day.

Having said all that, about all I can do is hop on the TjMax=100C bandwagon and then go back and do some more testing to see if this is reasonable across the entire temperature range or not.

Here's version 2.75 which incorporates the new IDF TjMax values for 45nm as well as similar changes to TjMax for 65nm.

http://www.fileden.com/files/2008/3/...alTempBeta.zip

As I previously mentioned, if TjMax really is 100C then some 45nm processors are going to have to be able to correct for a large amount of sensor error at idle. The calibration formula has been overhauled and it should provide more adjustability especially for processors that are reading too high. My original formula was designed early on in the 45nm era before some of these Quads started showing up with the Extreme sensor issues.

The formula is completely different so previous calibration factors will need to be adjusted. The range has been opened wide up from -9.9 to 9.9. Some users will misuse and mis-calibrate their CPUs trying to correct for stuck sensors, which is wrong, but other users may need to be able to do a significant correction so I won't stand in their way.

I did a few custom calibrations for users with really screwed up sensors so if you want to try out this new method then you'll need to contact me again and I'll see what I can do.

Have fun testing. This thread was getting kind of dead so at least we got something new to talk and argue about.

Edit: If you drag this new version into your old folder then it will read your old TjMax values. Open up the Settings screen and click on the Defaults button and it should set them to the new values. Either that or edit your INI file and delete the old TjMax values.

[WaterFlex]

I think the 2.75 shows the truth.

[Dua|ist]

unclewebb
I know you've probably got other stuff to do, but still a custom UI for dual-cores would be great to see. Just something like this:

Buttons could be still placed in one row if you could replace the text on them with some icons.
And BTW the new calibration algorithm seems to be way more flexible than previous one. I even managed to achieve same idle and almost same load temps on both cores (my chip has ~8-10 C difference between cores, no matter idle or full load). Once again thanks for another improvement!

[unclewebb]

Dua|ist: I've been thinking about creating a Dual Core specific version for a long time now and something very similar to what you've shown is in the near future for RealTemp. I just had to get the basics done first.

I'll be combining the information in the present two info windows at the top into one info window or maybe one line and one info window so there is enough room and I will be keeping my love it or hate it toggle button. It will be long and a little odd looking and use the same 4 buttons for consistency but when reduced to Mini Mode it will be much better looking for people with Dual Cores.

I'm glad to hear that the new calibration factors are working for you. Intel has not provided the user community with much guidance here so I've simplified my calibration formula based on the tiny scraps of info that Intel has thrown our way. If this formula helps 45nm processors produce some more believable looking numbers and everyone is more or less happy then I plan to publish exactly what I'm using so other software can follow along if they like.

Intel has now specifically stated that changes in the data coming from the thermal sensors does not happen at the exact same rate as changes in the core temperature. Temperature monitoring software needs to provide users with a way to adjust for the less than perfect sensors that Intel is using. I hope the competition doesn't continue to ignore this issue but they might. I figure if I release my formula as open source then everyone can get on the same page if they want to.

I just need some more user feedback from people like JohnZS who has a nice collection of random sensors on his 45nm Quad.

[eva2000]

The question I'm curious about is how does Intel determine TjMax? Is it an accurately measured value or is it an assumed value? Do they have a device that can accurately measure the core and determine that it is at 100C or do they measure 95C on the IHS of a finished unit and then say something like, "the core is probably about 5C hotter so we'll add on 5C and call it TjMax=100C." That's a serious question and one we'll never know the answer of.

I guess I'm not 100% convinced that switching RealTemp to TjMax=100C is going to result in more accurate temperatures. There was more important information left out of the IDF presentation than included in it. The only thing I'm really happy about is that I didn't spend a $1000 + travel to get my hands on that presentation. My rant for the day.

Having said all that, about all I can do is hop on the TjMax=100C bandwagon and then go back and do some more testing to see if this is reasonable across the entire temperature range or not.

Here's version 2.75 which incorporates the new IDF TjMax values for 45nm as well as similar changes to TjMax for 65nm.

http://www.fileden.com/files/2008/3/...alTempBeta.zip

As I previously mentioned, if TjMax really is 100C then some 45nm processors are going to have to be able to correct for a large amount of sensor error at idle. The calibration formula has been overhauled and it should provide more adjustability especially for processors that are reading too high. My original formula was designed early on in the 45nm era before some of these Quads started showing up with the Extreme sensor issues.

The formula is completely different so previous calibration factors will need to be adjusted. The range has been opened wide up from -9.9 to 9.9. Some users will misuse and mis-calibrate their CPUs trying to correct for stuck sensors, which is wrong, but other users may need to be able to do a significant correction so I won't stand in their way.

I did a few custom calibrations for users with really screwed up sensors so if you want to try out this new method then you'll need to contact me again and I'll see what I can do.

Have fun testing. This thread was getting kind of dead so at least we got something new to talk and argue about.

Edit: If you drag this new version into your old folder then it will read your old TjMax values. Open up the Settings screen and click on the Defaults button and it should set them to the new values. Either that or edit your INI file and delete the old TjMax values.

maybe time to close this thread and start a new one for 2.75 and up ?

[WaterFlex]

maybe time to close this thread and start a new one for 2.75 and up ?

I`m agree

[RealTelstar]

unclewebb
I know you've probably got other stuff to do, but still a custom UI for dual-cores would be great to see. Just something like this:

I love it!

[BeastNotro]

I say 95C for 65nm is the true TJMAX ... not 100C

[gtj]

I say 95C for 65nm is the true TJMAX ... not 100C

Actually, it may be 92.

The Intel Bone Trail and Bad Axe 2 motherboards come with Andigilog aSC7621 sensor chips that are wired to the processor PECI bus. So not only do they read the standard diode but they also get what the processor is reporting on the bus. If you check Intel's presentation from the other day you'll see how that relates to the DTS temps.

Anyway, one of the temps reported on the bus SEEMS to be the TjMax. For my E6700s and Q6600s, it's always 92 which fits perfectly with what the diode and DTS temps report. For instance, on my Q6600 crunchers that are running full blast 24x7, the diode temp is steady at 52c, the DTS from the MSR reads 40, PECI reports processor temp as -40 and the TjMax as 92. If I let the machine idle for a while, the diode drops to 34, DTS via MSR reads 58 and PECI reports -58 for the processor temp and .... 92 for the TjMax.

On my Q9300, PECI reports TjMax at 84 and the math works exactly as it does for the E6700s and Q6600s.

I'm going to test my E6850 and E8500 tonight.

[rge]

Tjmax is not software readable per intel on current cpus. The E6850, Q6600 GO, and E8400 all measure 95C at IHS when tjmax is reached (so tjmax can not be lower than 95). Given E8400 tjmax is 100 when IHS 95C, it would be reasonable to use the same gradient and thus tjmax for all that measure 95C, including Q6600 GO and E6850.

I was expecting Tjmax to be 95-97 from IHS testing, but given intel is the source and only talking about few C now (not like the way out 105 tjmax on E8400) hard to argue without a way to directly measure the core.

[JohnZS]

Dua|ist:
I just need some more user feedback from people like JohnZS who has a nice collection of random sensors on his 45nm Quad.

Hi Unclewebb
I have downloaded and installed the latest 2.75 beta (after removing the old one first)

I had a go at calibrating my extreme sensors, the ambient temperature is a bit cooler today, (instead of 23C it is 19/20C) hence slightly lower idle temps.



I will now do some stress testing to see how they behave when the core gets hot.

UPDATE

Here we are some nice load temps from Prime95


Thanks
John